Tilburg University
The actual structure of eBay’s feedback mechanism and early evidence on the effect of
recent changes
Klein, Tobias; Lambertz, C.; Spagnalo, G.; Stahl, K.O.
Published in:
International Journal of Electronic Business
Publication date:
2009
Link to publication
Citation for published version (APA):
Klein, T. J., Lambertz, C., Spagnalo, G., & Stahl, K. O. (2009). The actual structure of eBay’s feedback
mechanism and early evidence on the effect of recent changes. International Journal of Electronic Business,
7(3), 301-320.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright, please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Download date: 18. jun. 2017
Int. J. Electronic Business, Vol. 7, No. 3, 2009
301
The actual structure of eBay’s feedback mechanism
and early evidence on the effects of recent changes
Tobias J. Klein*
Department of Econometrics and OR,
Tilburg University,
P.O. Box 90153, 5000 LE Tilburg, The Netherlands
Fax: +31 13 466-3280
E-mail: [email protected]
*Corresponding author
Christian Lambertz
Department of Economics,
University of Mannheim,
68131 Mannheim, Germany
Fax: +49 621 181-1874
E-mail: [email protected]
Giancarlo Spagnolo
Faculty of Economics,
Department of Economics and Institutions,
University of Rome ‘Tor Vergata’,
Via Columbia 2, 00133 Rome, Italy
Fax: +39 06 2020500
E-mail: [email protected]
Konrad O. Stahl
Department of Economics,
University of Mannheim,
68131 Mannheim, Germany
Fax: +49 621 181-1874
E-mail: [email protected]
Abstract: eBay’s feedback mechanism is considered crucial to establishing and
maintaining trust on the world’s largest trading platform. Yet, there is
confusion among users about its exact institutional details, which changed
substantially in May 2007. Most importantly, buyers now have the possibility
to leave additional, anonymous ratings on sellers on four different criteria.
We provide a thorough description of the institutional details of eBay’s
feedback mechanism, including those changes. Then, we provide first
descriptive evidence on the impact of those changes on rating behaviour.
Copyright © 2009 Inderscience Enterprises Ltd.
302
T.J. Klein et al.
Keywords: eBay; reputation mechanism; strategic feedback behaviour;
informational content; reciprocity; fear of retaliation; electronic business.
Reference to this paper should be made as follows: Klein, T.J., Lambertz, C.,
Spagnolo, G. and Stahl, K.O. (2009) ‘The actual structure of eBay’s
feedback mechanism and early evidence on the effects of recent changes’,
Int. J. Electronic Business, Vol. 7, No. 3, pp.301–320.
Biographical notes: Tobias J. Klein is an Assistant Professor at Tilburg
University and is also affiliated with CentER, IZA, Netspar, and TILEC.
He obtained his PhD from the University of Mannheim. His research interests
are industrial organisation, policy evaluation, the economics of aging, and
various other topics in applied microeconomics.
Christian Lambertz is a PhD student at the University of Mannheim. He wrote
his undergraduate thesis on eBay and also some of his dissertation will consist
of research that is related to this paper.
Giancarlo Spagnolo is Professor at the University of Rome ‘Tor Vergata’.
He obtained his PhD from the Stockholm School of Economics. He is also
affiliated with SITE – Stockholm School of Economics, C.E.P.R., and
ENCORE. He has published, amongst others, in the RAND Journal of
Economics, the Journal of Economic Theory, the European Economic Review,
and the Journal of Money, Credit, and Banking.
Konrad O. Stahl is Professor at the University of Mannheim. He obtained his
PhD from the University of California, Berkeley. He is also affiliated with
C.E.P.R., CESifo, and ZEW. He has published numerous papers, amongst
others in the Bell Journal of Economics, the Journal of Industrial Economics,
the RAND Journal of Economics, and the Journal of Public Economics.
1
Introduction
EBay is the largest market ever to exist in terms of number of participants. It brings
together about 83 million active users. In 2007, the number of listings exceeded
2.3 billion, and eBay’s gross merchandise volume amounted to more than 59 billion
US dollars.1 This great success is often attributed to eBay’s feedback mechanism, which
is argued to foster trust in the platform (see, e.g., Resnick and Zeckhauser, 2002).
Trading on electronic platforms involves a particularly wide room for opportunistic
behaviour on both sides of the market. Anonymity and distance allow sellers to cheat
on the quality of the good. Likewise, buyers can be dishonest concerning their payment
behaviour (Lin et al., 2007).2
The need to enforce agreements and foster trust for exchange amongst strangers is not
limited to eBay. North (1990, 1991) argues that in general institutions play a major role
in reducing uncertainty by establishing a stable structure to human interactions.
Institutions consist of both informal constraints, such as traditions, and formal rules
that provide the incentive structure in an economy. On eBay, an escrow service,
the availability of buyer insurance, and the feedback mechanism mitigate the thread
of opportunistic behaviour. The escrow service is relatively expensive and typically
only used for purchases of $500 and more.3 Buyer insurance is provided as part of the
The actual structure of eBay’s feedback mechanism
303
Paypal payment service and is an effective way of buying insurance given that the seller
offers this payment option and the buyer is willing to pay the fees which amount
to about 3% of the transaction value.4 In this paper, we concentrate on the feedback
mechanism and investigate its ability to discipline the transaction parties by providing
the right incentives.
As long as feedback reflects the trading partners’ experience in a transaction,
accumulated feedback provides valuable information for potential trading partners.
Thereby, the feedback mechanism potentially disciplines the agents so that the
aforementioned forms of opportunism are attenuated by the threat that, if an agent
misbehaves today, she will receive bad feedback and will therefore be avoided by other
traders in the future. Public statements by eBay emphasise this (desired) effect of the
feedback mechanism.5
On eBay, both the seller and the buyer of an object can rate each other after a
transaction. It has been argued that this reciprocity might lead to strategic reciprocation
and retaliation: Leaving a positive rating could be driven by expectations to receive a
positive rating in return. This is consistent with the observation that usually a positive
rating is accompanied with an unrealistically favourable text comment such as
“Great transaction. A++++++ Seller”. At the same time agents dissatisfied with their
trading partners could refrain from leaving negative feedback at all, as they anticipate
revenge. Both forms of behaviour would bias the reputation index towards more positive
evaluations (Fehr and Schmidt, 1999; Resnick and Zeckhauser, 2002).6
The informational content of any user’s feedback record depends on the incentives
to report truthfully rather than strategically, that is, on the design of the feedback
mechanism. For example, if feedback is reciprocal and traders are hesitant to report a bad
transaction because they fear revenge from their trading partner, then a fixed length of the
time period in which feedback can be left would allow for leaving a negative rating
at the very last moment of this period. By ‘sniping’ negative feedback – i.e., leaving
it so late that the other party cannot react any more – users would be safe from retaliation.
Retaliating against deserved negative feedback and reciprocating positive ratings
may be used to build a reputation of being an imitator who always replies strategically
to a positive feedback with a positive one, and to a negative feedback with a negative
one. Such behaviour is in principle observable to other users on eBay as the feedback
a user gets and the replies she leaves can be inferred from her feedback record. It may be
valuable for a seller to be known as an imitator because it encourages her trading partners
to give a positive feedback and discourages them from giving a negative one. EBay even
sells a service to sellers allowing them to automatically reciprocate positive feedback.7
At the same time, if potential trading partners are fully aware of a seller being an imitator,
some of them will probably abstain from participating in the auction, knowing that a
negative first feedback would always be retaliated.
The fear of retaliatory negative feedback is regularly expressed in discussion
boards and newsgroups on eBay, and users indeed seem to be hesitant to leave
negative first ratings. Statements similar to the following from a buyer can be found
in many forums:8
“In the past I’ve not left any neg[ative] feedback as I’m afraid of revenge
feedback that’ll paint me as a bad trading partner … the dodgy seller ends up
with getting away with it just to rip someone else off.”
304
T.J. Klein et al.
The idea to leave negative ratings so late that the other party cannot react is often put
forward to solve the dilemma:9
“The secret … is to wait until the 90 day feedback period is nearly up and then
zap em w[ith a] negative feedback when they only have a few hours remaining
to respond … That way they can’t retaliate … This only wor[ks] if you are able
to hold a grudge for 90 days …”
Accordingly, McDonald and Slawson (2002) note that “some users attempt to avoid
retaliatory negative feedback by ‘sniping a negative’”. Auctionhawk, a company
specialised on offering services around eBay, even developed and advertised a service,
for payment, to give feedback in the last minute.10 One important contribution of this
paper is to show that ‘feedback sniping’ is in fact technically impossible on eBay, as will
be explained below.
Reputation is of central importance in electronic markets and the functioning of a
feedback mechanism depends on its design. In this paper, rather than focusing on the
effects of reputation, for example on prices or the probability of selling, we provide
detailed information on the precise design of the feedback mechanism on eBay, on which
there appears to be substantial confusion, and on how it changed in May 2007. Important
characteristics are the sequential nature of the feedback, the ending rule for the time
feedback can be left, the possibility to mutually withdraw ratings, and the additional,
anonymous seller ratings that can now be left by buyers.
Until the end of April 2007, a user’s reputation on eBay consisted of all ratings
received from his trading partners, buyers or sellers, on past transactions. In May 2007
the system changed. Since then, buyers have the possibility to leave additional ratings
on sellers, one to five stars, reflecting their satisfaction with the accuracy of the auction
listing, communication, shipping speed and shipping charges. These ratings can be left
anonymously and are summarised in additional summary statistics. After describing
the change, we provide descriptive evidence that is obtained from recently collected
data. The results suggest that the changes are a useful means to increase the likelihood
that dissatisfaction is actually communicated to the community.
The literature on strategic aspects of rating behaviour is sparse. Dellarocas (2006) and
Mailath and Samuelson (2006) discuss reputation building mechanisms in general.
Dellarocas et al. (2006) provide a practical guidance to design them. In the theoretical
literature on reputation, it is typically assumed that either the reputation bearer’s
behaviour is publicly observed by potential future trading partners, or that privately
observed behaviour is truthfully communicated from one market participant to another,
e.g. through “word of mouth communication”. An exception is Lippert and Spagnolo
(2006) where incentives to pass on information within a network of relationships are
analysed as well. However, there is no empirical research on this because word of mouth
communication is rather hard to document.
Electronic markets, especially their publicly observable feedback records, offer
economists the important chance to empirically analyse both the way reputation
as a collective good is built and how it acts, albeit in a somewhat special (anonymous
and public) environment. Resnick and Zeckhauser (2002) were among the first to provide
a descriptive analysis of rating behaviour on eBay. They report that, in their data,
sellers commented on buyers 60.6% of the time and buyers on sellers 52.1% of the time.
Ratings left by sellers and buyers were positive in 99.1% and 98.1% of the cases,
respectively. Furthermore, they find a high correlation between first and second
The actual structure of eBay’s feedback mechanism
305
ratings. They interpret this as evidence for reciprocity and retaliation. Other papers
acknowledge the possibility of strategic rating behaviour but focus on the effects
of reputation on selling prices and the probability of selling a good. The effects of
seller reputation on prices and the probability of selling the object are usually found to be
negligible or positive. See, for example, Melnik and Alm (2002), Bajari and Hortaçsu
(2003), Cabral and Hortaçsu (2006), Livingston and Evans (2004), Lucking-Reiley et al.
(2007), and Houser and Wooders (2006). In contrast, Anderson et al. (2007) focus on
typical listing strategies employed by occasional vs. large scale sellers and study the
effects of these strategies on prices and probability of sale. See also Bajari and Hortaçsu
(2004) as well as Resnick et al. (2004) for an overview. Reichling (2004) presents
descriptive evidence on rating behaviour and argues that users have an incentive to leave
untruthful ratings. Cabral and Hortaçsu (2006) provide a model of reputation building
on eBay but maintain the assumption that rating behaviour is truthful. Dellarocas and
Wood (2008) estimate a model of rating behaviour on eBay which is based on the
assumption that rating behaviour, if users decide to rate, is truthful. This precludes
the possibility that retaliative untruthful negative feedbacks are left. Using their model
they infer that buyers are satisfied with a transaction in about 80% of the cases whereas
sellers are satisfied in about 85% of the cases. These numbers compare to about 99%
positive feedbacks that are left on eBay.
The remainder of the paper unfolds as follows. Section 2 gives a detailed analysis
of the ‘classic’ part of the feedback mechanism. Section 3 describes the May 2007
changes to the system and the newly introduced reputation measures. In Section 4
we provide first empirical evidence on the effect of these changes. We conclude
in Section 5, suggesting additional simple ways to discourage opportunistic
feedback giving, and to improve on agents’ incentives to truly express their satisfaction
or dissatisfaction on eBay.
2
The ‘classic’ eBay feedback mechanism
When an auction ends or an item is sold, eBay sends a notification to the seller and
the buyer who should then contact each other and arrange payment and shipping.
As an intermediary, eBay assumes no responsibility in the process after the auction
has ended, and only gets involved if a dispute arises.
As soon as the auction is over (or the item is sold, if not through an auction),
both the seller and the buyer can leave feedback on their trading partner regarding
this transaction. Neither party is required to leave feedback, but is actively encouraged
by eBay to do so.
A feedback is a ‘positive’, ‘neutral’, or ‘negative’ rating accompanied by a textual
comment. When a rating is left, it is immediately observable to the counterpart and to the
community. The ratings a user receives are used to calculate his ‘feedback score’.
Generally, a positive rating increases the user’s feedback score by one point, a negative
rating decreases it by one point, and a neutral rating leaves it unchanged. A special
rule applies if a user interacts repeatedly with the same trading partner and receives more
than one rating from her: Then, the balance of these ratings is calculated. If the balance is
positive, the user’s score increases by just one point, if it is negative, the score decreases
306
T.J. Klein et al.
by just one point. Thus, even if users rate each other repeatedly, a member cannot affect
another member’s feedback score by more than one point.
The ‘feedback score’ is the most prominent indicator of a member’s reputation
on eBay and shown in parentheses next to his user ID wherever it is displayed
on eBay, no matter in what context. When the user acts as a seller – that is, on every item
page –, in addition to his feedback score, the percentage of positive feedback amongst
all positive and negative ratings she has received is reported. More detailed information
on any user, seller or buyer, as well as several summary statistics are available in the
‘feedback profile’ that eBay provides for every member. This profile includes a record
of all feedback that the user has ever received from or given to other members.
Members can make their feedback profile ‘private’. In that case, other users can
see only summary statistics of the feedback record. However, while his feedback
profile is private, a user is prohibited from selling on eBay. The ratings in a feedback
profile can be sorted so that only ratings received as a seller or ratings received as a buyer
are displayed.11
2.1 Last minute feedback?
With the reciprocity of the system, a detail of crucial importance is the ‘ending rule’
of the period in which trading partners can post their feedback (hereafter ‘feedback
period’). When retaliation is possible it is important whether this period is of fixed
or stochastic length.
We have argued that dissatisfied participants may be (rightly) concerned that
if they post a justified negative feedback the trading partner could retaliate with a
non-justified negative. This could only be avoided if it were possible to leave a rating
in the ‘last minute’ of a deterministic feedback period. Then, the trading partner would
have no time to retaliate.12 In the opposite case, i.e., without a fixed ending time of the
feedback period, the trading partner may retaliate so that many users will find it
unattractive to leave a first negative feedback. Consequently, if truthful reporting is a
welfare concern, the presence of a ‘last minute’ is desirable in the context of feedback.
While eBay only guarantees that feedback can be left within 90 days after an item
is sold, it seems to be a widespread perception that feedback cannot be left after this
90-day period. Examining our data base and again the structure of the eBay platform
we found, instead, that feedback can in fact be left as long as the auction details are
available. Moreover, if a first feedback is left, the party that received the rating has the
opportunity to reply for at least another 90 days. This is because whenever feedback
is left for a particular transaction, the rating is recorded in the feedback profiles
of the member that leaves and of the member that receives the rating. From there
the rating is linked to the item details for another 90 days – not after the end of the
auction, but starting with the time the first feedback has been left. The eBay system is
built in a way so that the details will be available for at least another 90 days. As a result,
the receiver of a first feedback will always have time to reply to this rating.
In light of the considerations about feedback retaliation and truthful reporting,
it is surprising that the end of the feedback period is stochastic on eBay, and in fact
automatically extended after the first rating. In fact, the current real ending rule
of the feedback period on eBay is similar to the one of Amazon for auctions.13 With this
structure, the eBay feedback mechanism in principle discourages truthful negative
The actual structure of eBay’s feedback mechanism
307
ratings by the potential first feedback giver, by giving the opponent enough time
to retaliate.
2.2 Mutual feedback withdrawal
Another important feature of the mechanism in this context is the possibility to
mutually withdraw feedback. Once recorded, a rating cannot be unilaterally removed.
However, what probably not all users have noticed, feedback can be withdrawn
if both parties agree to it.14 Because of this option, the receiver of a first negative
feedback has an incentive to strategically reply with a negative feedback in order
to induce her trading partner to agree to a mutual withdrawal of ratings. She might
consider doing this even if she is in fact happy with the behaviour of her trading partner
in the underlying transaction. If both trading partners then agree on a feedback
withdrawal, again, no negative feedback would be observed. If an eBay user is not fully
aware of this fact, she might overestimate the informational content of feedback records
(Jin and Kato, 2007; Resnick et al., 2004).15
To summarise, key elements of the classic feedback mechanism are the following:
1
feedback is immediately observable
2
there is always a time window allowing a trading partner to react to a rating as long
as she has not entered one yet
3
feedback can be withdrawn upon the mutual consent of both trading partners.
Figure 1 contains a state chart representing these key elements. Each circle in the
graph represents a state and each arrow a transition from a state into another state
or into itself. Such a transition, and if only from a state into itself, happens at every
instant in time, i.e., at every moment, we move along some arrow in the state chart.
We can describe the feedback game using the state chart: We enter from the left.
We are in the state in which nobody has left feedback so that it can still be left
by both. In the next instance of time, either no one rates and the feedback period is not
over, or only one of the two parties rates, or both rate each other simultaneously, or the
feedback period is over. Depending on the actions of the players we transition
into another state.
The last (grey shaded) state is always the payoff state. Payoffs are to be understood
as expected payoffs from future transactions on eBay, conditional upon the feedback
outcome. Depending on the history, either no feedback has been left, or one or two
ratings have been left without being withdrawn – the usual case –, or ratings have been
left and were withdrawn thereafter.16
The dashed part of the graph represents the misperception of the existence of a last
minute of the feedback period that was discussed above. To be more specific, the
misperception is that after 90 days there is a transition into a ‘last minute’ state in which
the trading partner cannot react to a rating. Most importantly, the chart (without the
dashed part) shows that once a first feedback is left, in fact, the trading partner always has
the opportunity to react with a second feedback.
308
Figure 1
T.J. Klein et al.
State chart (see online version for colours)
The actual structure of eBay’s feedback mechanism
3
309
May 2007 changes
As already mentioned, the eBay feedback mechanism changed substantially in May 2007.
Since then buyers can, in addition to the ‘classic’ feedback, leave detailed ratings
on a seller regarding four different criteria: item as described, communication,
shipping time, and shipping and handling charges. A buyer can award the seller
one to five stars on each of these aspects. Detailed seller ratings are optional, and it is left
to the buyer’s discretion whether she wants to rate the buyer in any of these areas.17
If a buyer wants to leave detailed ratings, she must do so at the same time as she leaves
the classic ‘overall’ feedback with textual comment. Classic feedback can be left without
providing detailed ratings, but not the other way around. Concerning their substance,
however, overall feedback and detailed ratings need not be connected in any way.
For example, the buyer could leave a positive classic feedback, and at the same time give
only one star on each of the four criteria. The two systems co-exist to allow for a smooth
transition from the old to the new system. This is necessary because users have built up
their feedback records over many years and classic feedbacks cannot be converted
into detailed ones. However, as more and more detailed ratings will be collected classic
feedbacks are likely to become less and less important.
There are two important ways in which the detailed ratings are different from
the classic ratings. First, the granularity of the type of rating that can be given is
different. In the classic system, ratings consist of a positive, neutral or negative overall
assessment and every mark that is not positive is usually considered as a negative mark,
see, e.g., Resnick and Zeckhauser (2002). The detailed ratings consist of 1-5 stars along
four dimensions. This allows users to be more precise in their assessment of their trading
partners’ performance (Dellarocas, 2005). Moreover, eBay attaches meanings such as
‘satisfactory’ to the different star ratings, see Table 1 for details.
Table 1
Meaning of stars
1 star
2 stars
3 stars
How accurate was the
item description?
Very
inaccurate
Inaccurate
Neither inaccurate Accurate
nor accurate
Very
accurate
How satisfied were you
with the seller’s
communication?
Very
unsatisfied
Unsatisfied
Neither unsatisfied Satisfied
nor satisfied
Very
satisfied
Neither slowly nor Quickly
quickly
Very
quickly
How quickly did the seller Very slowly Slowly
ship the item?
How reasonable were the Very
Unreasonable Neither
shipping and handling
unreasonable nor
unreasonable
charges?
reasonable
4 stars
5 stars
Reasonable Very
reasonable
In the empirical analysis we multiply the numbers of stars by 100.
Second, the new detailed seller rating is anonymous. For each criterion, only the average
of all ratings is given, and a seller must receive at least ten ratings before the average is
reported in his feedback profile. eBay stresses that “sellers will not be able to see the
detailed seller ratings you’ve given them”,18 and when actually leaving detailed ratings,
buyers are notified again that “sellers will not see your individual ratings” and that
310
T.J. Klein et al.
“only the average of all buyer ratings can be seen by the seller”. In other words, buyers
are assured that, with the new detailed seller ratings, they are safe from retaliation.
As with the classic feedback where a specific buyer can affect a seller’s score by only
one point, if a buyer leaves detailed ratings on the same seller for different transactions,
only the average of her ratings enters the seller’s statistic for that criterion. In contrast
to the classic feedback score, averages for detailed ratings will be calculated over the
preceding 12 months only.19
At the time eBay introduced detailed seller ratings, it also introduced minor changes
to the other part of the feedback system, in particular to the way in which ‘classic’
ratings are displayed. Before, apart from the auction number, only the rating – ‘positive’,
‘neutral’, or ‘negative’ –, the textual comment, the feedback giver and the date and time
when the rating was left were displayed. For at least 90 days after the end of the auction
there was a link to the auction details and users had to follow that link to learn more
about the transaction. Today, in addition to this information, the item title and the price at
which it was sold are also reported in the feedback profile. Potential buyers can thus see
at a glance from what kind of item a recent rating stems, and whether it was given for a
high value or on a low value transaction.
To summarise, we predict that users are more inclined to report dissatisfaction in the
new system. This is for two reasons. First, the change in granularity of the rating scale
makes it less likely that the best rating is given, simply because the scale is finer. Second,
the incentive to rate truthfully, if unsatisfied, is considerably stronger in the new system
because ratings are anonymous.
4
Empirical results
The data we use were collected in several steps. First, in May 2007, we obtained
the usernames of, respectively, the last 3,000 users that were selling in the five categories
listed in Table 2.20 This is a representative sample of those who sold at least once in those
five categories. Then, on the 1st of June, July, August, and September, we collected their
feedback overview pages, one per user. This gave us detailed information on user
characteristics such as the overall percentage of positive marks received over their life as
an eBay user. Most importantly it gave us information on ratings received in the five
months prior to the change and in the first four months after the change.21 From this raw
data we construct percentages for the classic ratings in order to compare them to the
average number of stars obtained in the new part of the feedback mechanism.
Table 2
Categories
Home > All categories > Computers and Networking > Laptops, Notebooks
Home > All categories > Consumer Electronics > Apple iPod, MP3 Players
Home > All categories > Toys and Hobbies > Model RR, Trains
Home > All categories > Collectibles > Trading Cards
Home > All categories > Home and Garden > Food and Wine
These categories are from the USA eBay system.
The actual structure of eBay’s feedback mechanism
311
Table 3 contains summary statistics. They are reported for our core sample consisting
of users for which detailed ratings are available.22 They show that there are both some big
players with large feedback records but also smaller ones. The distribution of feedback
scores is highly skewed with a mean of 1,998 and a median of 616, see Table 4.
About 29% of these sellers are power sellers. Interestingly, in the first four months
after the introduction of the new system a seller in our sample received 317 classic
ratings and 163 detailed ratings on average. This reflects the fact that detailed ratings can
only be received once users act as sellers.
Table 3
Summary statistics
Variable
Obs.
Mean
Std.
Min.
Max.
Duration membership in years
3,704
4.71
2.79
0.27
11.35
Member is power seller
3,704
0.29
0.45
0.00
1.00
Member has ‘About me’ page
3,704
0.21
0.40
0.00
1.00
Member has store
3,704
0.29
0.45
0.00
1.00
Received ratings 1/12/2006–30/4/2007
3,704
424.52 1,279.44
0.00
29,913.00
Positive
3,704
420.17 1,258.00
0.00
29,371.00
Neutral
3,704
2.46
15.58
0.00
565.00
Negative
3,704
1.89
14.91
0.00
599.00
Neutral or negative
3,704
4.35
29.56
0.00
1,164.00
Positive
3,600
0.99
0.02
0.50
1.00
Neutral
3,600
0.00
0.01
0.00
0.33
Negative
3,600
0.00
0.01
0.00
0.50
Neutral or negative
3,600
0.01
0.02
0.00
0.50
Percentage ratings 1/12/2006–30/4/2007
Received ratings 1/5/2007–31/8/2007
3,704
317.02
805.75
9.00
16,376.00
Positive
3,704
312.87
790.78
6.00
16,099.00
Neutral
3,704
2.28
10.59
0.00
189.00
Negative
3,704
1.88
9.71
0.00
229.00
Neutral or negative
3,704
4.15
19.57
0.00
390.00
Positive
3,704
0.99
0.03
0.43
1.00
Neutral
3,704
0.01
0.01
0.00
0.21
Negative
3,704
0.01
0.02
0.00
0.49
Neutral or negative
3,704
0.01
0.03
0.00
0.57
Item description
3,697
478.57
17.22 240.00
500.00
Communication
3,686
472.92
21.30 260.00
500.00
Shipping time
3,691
467.24
27.07 280.00
500.00
Shipping charges
3,683
458.19
21.84 330.00
500.00
Percentage ratings 1/5/2007–31/8/2007
Detailed ratings 1/5/2007–31/8/2007 (from 100 to 500)
312
T.J. Klein et al.
Table 3
Summary statistics (continued)
Variable
Obs.
Std.
Min.
Mean number of detailed ratings 1/5/2007–31/8/2007
3,704
162.63
492.94
10.00
13,955.50
Item description
3,697
163.60
495.46
10.00
14,016.00
Communication
3,686
163.00
492.53
10.00
13,875.00
Shipping time
3,691
163.26
494.00
10.00
13,990.00
Shipping charges
3,683
163.10
493.32
10.00
13,941.00
1.00
148,980.00
48.60
100.00
Member feedback score
3,704
Overall percentage positives
3,704
Mean
1,998.01 5,647.45
99.48
1.50
Max.
The observational unit is a user who has offered a good in one of the categories listed
in Table 2 and for whom detailed ratings are available. Received ratings are classic
ratings which are either positive, neutral or negative. These ratings are not anonymous
and are directly observable to the trading partner. Detailed ratings are the ratings that
can now be left additionally to the traditional ratings. Here, 1 to 5 stars can be left,
respectively. We multiply the average number of stars received, per user, by 100.
The number of ratings differs across detailed and traditional ratings because detailed
rating are optional and can only be left for sellers. The member feedback score is the
number of unique users who left a net positive rating minus the number of unique users
who left a net negative rating. The overall percentage negative ratings is calculated from
all ratings the user has received.
Table 4
Percentiles of the distribution of feedback scores by availability of detailed ratings
Percentile
Detailed rating
1
5
10
25
50
75
90
95
99
No
0
1
3
15
60
189
446.5
705
1,650
Yes
26
59
99
229
616
1,646
7,612
24,714
4,199
Our data confirm the well known empirical finding that the overwhelming share of
classic ratings is positive. Interestingly, this share is declining once we compare the five
months prior to the introduction of the new rating possibilities to the first four months
after the system has been put in place. This decline is small but significant with a p-value
of 0. One interpretation of this is that the recent changes have motivated some users
to expose themselves to the risk of retaliation. It is an open question whether this can be
reconciled with rational behaviour.
Figure 2 contains box plots for the average number of stars received by the users
in our sample. Overall, these ratings are very favourable. In part this could be driven
by the fact that detailed ratings are only available for active players, namely those who
have received at least ten detailed ratings in their role as a seller over the first four
months the new system was in place. As we have started with 3,000 sellers in each of the
five categories this is only the case for 3,700 out of 15,000 sellers. Still, it is interesting to
see that the typical seller receives many ratings that are not of the best type,
5 stars. This is in contrast to the finding that almost all classic ratings are of the
best type, positive, and is well consistent with the arguments that have been made
in Sections 2 and 3.
The actual structure of eBay’s feedback mechanism
Figure 2
313
Box plots for detailed ratings
The horizontal axis is the average number of stars for a given user multiplied by 100.
The box is bounded on the left by the first quartile and on the right by the third
quartile. The observations outside the two whiskers are considered to be outliers
and are reported. The observation on the very left is the lowest observed value,
respectively.
Table 5 contains, by category and percentage positive ratings received within the same
period, sample means of the average number of stars received by the user. It illustrates
the dependence between the two measures and shows that even for users who have
received less than 95% positive feedbacks the detailed ratings are very good, e.g.,
on average between ‘accurate’ and ‘very accurate’ for the item description.
Table 5
Detailed ratings
Obs. Item description Communication Shipping time Shipping charges
100% feedbacks positive
At most 95% feedbacks positive
3,669
478.624
472.976
467.291
458.237
1,919
483.715
480.130
475.393
464.914
185
446.703
426.595
416.054
425.676
This table shows, in the columns, respectively, the condition on the percentage positive
feedbacks received, the number of observations and the average rating received across
users. The average number of stars is multiplied by 100. For example, 483.715 is the
average score for the item description among all users who received only positive ratings.
Reported for users for which all four summary statistics are reported.
Finally, Table 6 contains results from a regression of the average star rating on a rich set
of user characteristics. It illustrates that the number of stars depends positively on the
number of detailed ratings received, the percentage positive feedbacks received, and the
overall percentage positive feedbacks. It depends negatively on the feedback score and is
smaller for power sellers.
314
T.J. Klein et al.
Table 6
Regression results
Item description
Category notebooks
Category mp3 players
Category model rr trains
Category trading cards
Log of mean number of
detailed ratings
Communication Shipping time Shipping charges
–5.803
–6.069
–2.855
–8.883
(7.75)**
(7.21)**
(2.37)*
(8.65)**
–6.691
–5.708
–3.592
–10.939
(7.93)**
(6.01)**
(2.65)**
(9.44)**
–0.061
–3.310
–1.275
–2.140
(4.92)**
(1.69)
(1.98)*
(0.07)
–0.414
–1.559
–4.749
–0.109
(0.68)
(2.28)*
(4.85)**
(0.13)
1.287
1.454
2.567
0.071
(4.18)**
(4.20)**
(5.18)**
(0.17)
–391.187
–509.203
–697.092
–421.991
Percentage neutral ratings
(18.71)**
(21.69)**
(20.64)**
(14.73)**
Percentage negative
ratings
–40.905
–204.224
–192.535
–4.601
(3.60)**
(16.00)**
(10.02)**
(0.30)
Log feedback score
Overall percentage
positive feedbacks
Duration membership
Member is power seller
Member has ‘About me’
page
Member has store
–0.710
–1.661
–1.970
–0.586
(2.44)*
(5.09)**
(4.22)**
(1.47)
4.816
4.825
4.738
4.651
(410.77)**
(365.79)**
(251.89)**
(289.25)**
0.669
0.811
1.292
0.852
(6.26)**
(6.75)**
(7.52)**
(5.81)**
–0.384
–0.985
–0.996
–2.261
(0.62)
(1.43)
(1.01)
(2.68)**
0.223
1.852
0.842
1.585
(0.37)
(2.72)**
(0.87)
(1.91)
1.637
0.244
–0.883
0.390
(2.84)**
(0.38)
(0.95)
(0.49)
Obs.
3,697
3,686
3,691
3,683
R-squared
1.00
1.00
1.00
1.00
Absolute value of t statistics in parentheses
* Significant at 5%; ** significant at 1%
The columns contain coefficient estimates and corresponding t statistics that were
obtained from a regression of the respective dependent variable on the covariates listed
above. The dependent variable is the average number of stars received multiplied by 100.
We conclude from these first empirical results that the rating possibilities that were
recently added to the feedback mechanism tend to add valuable information. In particular,
about 99% of classic ratings are of the best type, positive, but this seems not to be the
case for the detailed ratings.23 In our previous discussion we have related this to the fact
that here, the fear of a retaliatory feedback does not play a role as the type of rating
cannot be inferred by the seller.
The actual structure of eBay’s feedback mechanism
5
315
Conclusions
EBay’s reputation mechanism is an important institution in an environment of impersonal
exchange where information about the attributes of the goods traded and the performance
of the agents involved is imperfect. Milgrom et al. point out that
“a good reputation can be an effective bond for honest behaviour in a
community of traders if members of the community know how others have
behaved in the past – even if any particular pair of traders meets only
infrequently.” (Milgrom et al., 1990)
Whether the system employed by eBay succeeds in this respect depends on its setup.
In this paper we have investigated the institutional details of both the classic part
of eBay’s reputation mechanism and its newly added features. It is well known that
the former allows for the immediate observation of feedbacks given by the trading
partners. Therefore, the choice of the timing of feedbacks and their type may be guided
by strategic considerations. Leaving a truthful negative feedback is a risky endeavour
because it may be retaliated untruthfully, and purely for strategic reasons. We have
argued that the existence of a deterministic last minute of the feedback period would
allow for negative first feedback giving without retaliation. However, closer scrutiny of
eBay’s reputation mechanism reveals that a trading partner has always enough time to
react to a first rating. This implies that a negative first feedback can never be given
without the fear of retaliation. One aim of this paper was to emphasise this point.
A direct consequence of this institutional design is that a first feedback giver,
if behaving opportunistically towards establishing a good feedback record, may
not give a negative feedback even if it were justified. Instead, she might wait
for her partner to give the first feedback. If both wait, no feedback is left at all
(Dellarocas and Wood, 2008).24 Conversely, leaving a positive feedback might be driven
by expectations on feedback reciprocity which induces the trading partner to react with a
positive mark. For this reason, many studies have argued that classic ratings do not
directly reflect the true performance of the participating agents. Importantly, this does not
apply to the recently added rating possibilities that allow buyers to unilaterally rate
sellers, as they ensure anonymity.25 The difference between the two systems has been at
the centre of the empirical analysis that has been presented in Section 4 of this paper.
One important finding is that whereas about 99% of the classic ratings are of the best
type, positive, this is not the case in the new system.
In addition to the changes that were made in May 2007 there may be scope
for further improvements of the classic part of eBay’s feedback mechanism using
easy-to-implement measures. In particular, our analysis suggests that to reduce concerns
for retaliation and to foster expression of deserved dissatisfaction, the revelation of
information should be delayed.26 Most importantly, feedback should only be
revealed to the trading partners and thus, the public, if no more feedback can be left.27
This could alternatively be done after a fixed period, or after both trading partners have
given their feedback. A direct consequence could be that less users might be willing to
leave a first (positive) rating because they cannot hope any more to thereby induce their
trading partner to leave a rating in return. However, this might be a desired effect as,
on average, the rating would be closer to the true satisfaction of the user
who left it. In addition, incentives to rate (e.g., reductions on future fees) could be
provided or ratings could be made (quasi-) mandatory by not allowing users to bid in
316
T.J. Klein et al.
another auction or add a new auction listing if they have not rated within a specified
time. An important point is that this change alone would potentially make things even
worse as long as the mechanism would still allow for the mutual withdrawal of feedback.
We therefore advocate to also remove this option. Otherwise, it tends to remain a
dominant strategy to always leave negative feedbacks in order to be able to renegotiate
after ratings have been revealed.
Finally, it can be argued that the performance of buyers, if asked to pay first,
is subject to little uncertainty. Either the full payment arrives in time, which is verifiable
by the seller, or it does not. Then, opportunism on the buyer’s side does not play a role.
To the contrary, sellers can misbehave on a variety of aspects of their performance.
Therefore, it may be worthwhile to limit feedback to buyers rating sellers as in Amazon
auctions (Dini and Spagnolo, 2006; Dellarocas et al., 2006).28 Such a change of rules is
likely to induce less feedback giving. However, by removing potentially very substantive
biases, it should increase the informational content of feedback records and therefore lead
to an unequivocal improvement in the allocation decisions taken in this increasingly
important market.
6
Future research
This paper provides a detailed discussion of the institutional details of eBay’s reputation
mechanism. We have focussed on the formal incentives that are provided to the trading
partners to communicate experiences truthfully to the community. More generally,
North (1990) points out that it is the combination of such formal incentives and other
informal constraints that induce agents’ behaviours. The same formal rules will result
in different outcomes when applied in different countries which differ in their culture.
As a global marketplace, eBay provides the opportunity to study the effects of its
(same) mechanism in different environments. An analysis in this direction would
certainly be valuable.
Besides, and more closely related to what has been done in this paper, it would
be interesting to conduct an analysis similar to ours with 2008–2009 data. One of the
empirical findings in this paper is that average detailed ratings for big sellers are very
positive, indicating that the reputation mechanism is effective in providing the right
incentives to those sellers. It would be interesting to see whether this also holds true for
small sellers for which those detailed ratings were not available in the data that we
collected as statistics are only displayed once 10 detailed ratings have been received.
Moreover, more detailed data could be used to investigate how many classic ratings a
seller has received as a seller and compare this to the number of detailed ratings that she
can only receive as a seller. By construction, there is always a classic rating for every
detailed rating, but not vice versa. One could infer from this how willing buyers are to
provide detailed ratings. This in turn would indirectly reveal how useful buyers find
detailed ratings for evaluating the performance of sellers.
A more direct way to infer this would be to study the effect of detailed ratings on the
probability of selling a good and the selling price since this is informative about the way
potential buyers interpret those summary measures. For this, data on the auction level
have to be collected, which is a somewhat tedious but worthwhile endeavour.
The actual structure of eBay’s feedback mechanism
317
Acknowledgements
We would like to thank Heski Bar-Isaac, Chris Harris, Ali Hortaçsu, Johannes Koenen,
Florian Müller, Axel Ockenfels, Klaus Schmidt, Henry Schneider, and Mike Ward for
stimulating discussions and comments on related work. Florian Hauber and Johannes
Koenen provided outstanding research assistance. We would also like to thank seminar
participants at the ESMT in Berlin and the University of Siena as well as conference
participants of the EARIE 2008 in Toulouse for their comments. Financial support from
the Deutsche Forschungsgemeinschaft through SFB/TR 15 is gratefully acknowledged.
Finally, we would like to thank the editor and two referees for their helpful suggestions.
References
Anderson, S.T., Friedman, D., Milam, G. and Singh, N. (2007) ‘Seller strategies on eBay:
Does size matter?’, International Journal of Electronic Business, Vol. 5, No. 6, pp.643–669.
Bajari, P. and Hortaçsu, A. (2003) ‘The winner’s curse, reserve prices, and endogenous entry:
empirical insights from eBay auctions’, RAND Journal of Economics, Vol. 34, No. 2,
pp.329–355.
Bajari, P. and Hortaçsu, A. (2004) ‘Economic insights from internet auctions’, Journal of Economic
Literature, Vol. 42, No. 2, pp.457–489.
Cabral, L.M.B. and Hortaçsu, A. (2006) The Dynamics of Seller Reputation: Theory and Evidence
from eBay, Mimeograph, New York University, New York, NY.
Dellarocas, C. (2006) ‘Reputation mechanisms’, in Hendershott, T. (Ed.): Handbook on
Information Systems and Economics, Elsevier, Amsterdam, NL.
Dellarocas, C. (2005) ‘Reputation mechanisms design in online trading environments with pure
moral hazard’, Information Systems Research, Vol. 16, No. 2, pp.209–230.
Dellarocas, C. and Wood, C.A. (2008) ‘The sound of silence in online feedback: estimating trading
risks in the presence of reporting bias’, Management Science, Vol. 54, No. 3, pp.460–476.
Dellarocas, C., Dini, F. and Spagnolo, G. (2006) ‘Designing feedback mechanisms for
e-procurement platforms’, in Dimitri, N., Piga, G. and Spagnolo, G. (Eds.): Handbook
of Procurement, Chap. 18, Cambridge University Press, Cambridge, UK, pp.446–482.
Dini, F. and Spagnolo, G. (2006) ‘Reputation mechanisms and electronic markets: economic issues
and proposals for public procurement’, in Thai, K.V., Araujo, A., Carter, R.Y., Callender, G.,
Drabkin, D., Grimm, R., Ejlskov Jensen, K.R., Lloyd, R.E., McCue, C.P. and Telgen, J.
(Eds.): Challenges in Public Procurement: An International Perspective, Academic Press,
Gent, BE.
Fehr, E. and Schmidt, K. (1999) ‘A theory of fairness, competition and cooperation’, Quarterly
Journal of Economics, Vol. 114, No. 3, pp.817–868.
Houser, D. and Wooders, J. (2006) ‘Reputation in auctions: theory, and evidence from eBay’,
Journal of Economics and Management Strategy, Vol. 15, No. 2, pp.353–370.
Jin, G.Z. and Kato, A. (2007) ‘Dividing online and offline: a case study’, Review of Economic
Studies, Vol. 74, No. 3, pp.981–1004.
Lin, Z., Li, D. and Huang, W.W. (2007) ‘Traders beware: an examination of the distribution
of eBay sellers’ online reputation’, International Journal of Electronic Business, Vol. 5, No. 5,
pp.499–517.
Lippert, S. and Spagnolo, G. (2006) SSE/EFI Working Paper in Economics and Finance No. 570,
Stockholm School of Economics, Stockholm, SE, available at http://swopec.hhs.se
Livingston, J.A. and Evans, W.N. (2004) Do Bidders in Internet Auctions Trust Sellers?
A Structural Model of Bidder Behaviour on eBay, Working Paper, Bentley College.
318
T.J. Klein et al.
Lucking-Reiley, D., Bryan, D., Prasad, N. and Reeves, D. (2007) ‘Pennies from eBay: the
determinants of price in online auctions’, Journal of Industrial Economics, Vol. 55, No. 2,
pp.223–233.
Mailath, G.J. and Samuelson, L. (2006) Repeated Games and Reputations: Long-Run
Relationships, Oxford University Press, Oxford, UK.
Mcdonald, C.G. and Slawson, V.C. (2002) ‘Reputation in an internet auction market’, Economic
Inquiry, Vol. 40, No. 3, pp.633–650.
Melnik, M.I. and Alm, J. (2002) ‘Does a seller’s reputation matter? Evidence from eBay auctions’,
Journal of Industrial Economics, Vol. 50, No. 3, pp.337–349.
Milgrom, P.R., North, D.C. and Weingast, B.R. (1990) ‘The role of institutions in the revival of
trade: the law merchant, private judges, and the champagne fairs’, Economics and Politics,
Vol. 2, No. 1, pp.1–23.
North, D.C. (1990) Institutions, Institutional Change and Economic Performance, Cambridge
University Press, Cambridge, UK.
North, D.C. (1991) ‘Institutions’, Journal of Economic Perspectives, Vol. 5, No. 1, pp.97–112.
Reichling, F. (2004) Effects of Reputation Mechanisms on Fraud Prevention in eBay Auctions,
Thesis, Stanford University, Stanford, California.
Resnick, P. and Zeckhauser, R. (2002) ‘Trust among strangers in internet transactions: empirical
analysis of eBay’s reputation system. The economics of the internet and e-commerce’,
in Baye, M.R. (Ed.): Advances in Applied Microeconomics, Elsevier Science, Amsterdam,
Vol. 11, pp.127–157.
Resnick, P., Zeckhauser, R., Swanson, J. and Lockwood, K. (2004) The Value of Reputation
on eBay: A Controlled Experiment, Working Paper, Harvard Kennedy School of Business.
Roth, A.E. and Ockenfels, A. (2002) ‘Last-minute bidding and the rules for ending second-price
auctions: evidence from eBay and Amazon on the internet’, American Economic Review,
Vol. 92, No. 4, pp.1093–1103.
Notes
1
See http://investor.ebay.com/common/download/download.cfm?companyid=ebay&fileid=192977
&filekey=08eaa22c-c31c-41ad-b4d3-0a954354566f&filename=ar2007.pdf (January 2009).
2
According to the Internet Crime Complaint Center (IC3) 2005 Internet Fraud Crime Report
“internet auction fraud was by far the most reported offence, comprising 62.7% of [97,076]
referred complaints”. See http://www.ic3.gov/media/annualreports.aspx (September 2006).
Likewise, the FTC reports that “internet auction fraud is on the rise, with an increasing number of
consumers complaining about sellers who deliver their advertised goods late or not at all,
or deliver something far less valuable than promised”. See the FTC’s “Top Ten Dot Cons” on
http://www.ftc.gov/bcp/conline/edcams/dotcon/auction.htm (February 2006). Lin et al. (2007) find
that on eBay, most complaints from buyers did in fact not refer to online fraud. Moreover, they
find that buyers could be responsible for online disputes more often than sellers.
3
The service is provided by a company that collects the money from the buyer and the good
from the seller and only then sends the money to the seller and the good to the buyer.
See http://pages.ebay.com/help/pay/escrow.html (January 2009).
4
Typically, if the seller accepts payments via Paypal, the buyer has to pay the extra fees.
See http://pages.ebay.com/help/buy/paypal-buyer-protection.html (January 2009) for Paypal’s
buyer protection.
5
eBay states that the feedback “comments and ratings are valuable indicators of your reputation
as a buyer or seller on eBay”, see http://pages.ebay.com/help/feedback/questions/feedback.html
(February 2006). Moreover, in the founder’s letter posted on February 26, 1996, Pierre Omidyar
claims that “some people are dishonest. Or deceptive … But here, those people can’t hide.
The actual structure of eBay’s feedback mechanism
319
We’ll drive them away”. See http://pages.ebay.com/services/forum/feedback-foundersnote.html
(February 2006).
6
This tendency to reciprocate may be due to behavioural components in agents’ decision making
processes, similar to the ones found by Fehr and Schmidt (1999), the attempt to build up a
reputation as a ‘reciprocator’ or ‘impersonator’ in order to discourage future negative ratings and
encourage positive ones – “the high courtesy equilibrium” of Resnick and Zeckhauser (2002) –, or
the combination of both motives.
7
The price for an online seller tool which includes this service is currently $15.99 a month, see
http://pages.ebay.com/sell/automation.html (February 2006) for a description.
8
Quote taken from http://ideas.4brad.com/archives/000018.html (February 2006). See the
newsgroup discussions on http://www.the-gas-station.com/messages.cfm?type=normal &thread_id
=49933&lastdays=2000&
and
http://community.auctionsniper.com/groupee/forums/a/tpc/f/
785608021/m/308108399/r/3721016131, for example (February 2006).
9
Quote
from
http://www.the-gas-station.com/messages.cfm?type=normal&thread_id=49933
&lastdays=2000& (February 2006).
10
See http://auctionbytes.com/cab/abn/y04/m08/i10/s01 (February 2006). A free reminder service
for “last minute feedback” is offered by UK Auction Watch at http://www.ukauctionhelp.
co.uk/remindme.php (February 2006).
11
On the German site, for example, users have the additional option to display only ratings from the
past month, or the past 6 or 12 months. For these periods, users can then choose to display
feedback by type, i.e., only positive, only neutral, or only negative – or only withdrawn – ratings.
This feature is not available on USA eBay; it is on other sites.
12
From a strategic viewpoint, this closely resembles “last minute bidding” in English auctions with
fixed ending time (Roth and Ockenfels, 2002), with a “last minute action” being exploited in both
cases in order to prevent opponents from reacting.
13
There, a ‘last minute’ bid prolongs the auction period automatically. See http://www.
amazon.com/gp/help/customer/display.html?ie=UTF8&nodeId=1161360 (September 2006).
14
eBay states that
“[a]fter both parties have agreed to withdraw the feedback, both parties will
have their feedback scores adjusted at the same time … eBay will add a note to
the feedback comment, saying that the feedback was mutually withdrawn … If
you haven’t left feedback for your trading partner and you go through the
Mutual Feedback Withdrawal process, you will no longer be able to leave
feedback for that transaction … You may only request Mutual Feedback
Withdrawal once for every feedback left … Members may initiate a request to
mutually withdraw feedback within 30 days of either person leaving feedback
or within 90 days of the transaction end date, whichever is later.”
Taken from http://pages.ebay.com/help/feedback/questions/mutual-withdrawal.html (September
2006).
15
Jin and Kato (2007) find in a field experiment that “at least some buyers” overestimate the
informational content of feedback score and “drastically underestimate the risk of trading online”.
Likewise, Resnick et al. (2004) question whether price premia, which they find, reflect a
reputation equilibrium, and should in fact not be observed in the data.
16
For ease of the exposition, we make the simplifying assumption that we can always enter the
feedback withdrawal process after at least one feedback has been left. In reality, every player may
initiate this only once, see footnote 14 for details. Note that only a subset of the users on eBay is
likely to be aware of the possibility of withdrawal.
17
Buyers cannot rate on ‘shipping time’ and ‘shipping and handling charges’ in motor vehicle
categories.
18
http://pages.ebay.com/help/feedback/detailed-seller-ratings.html (September 2007).
19
Note that there are (rare) situations in which the rating can be inferred. This is because the
average rating is presented accurate to a tenth of a star. For example, if a user has an average
320
T.J. Klein et al.
rating of 5 stars from ten ratings and receives an 11th rating he can infer it if it changes his
average rating, i.e., if it is at most 4 stars. However, in typical situations this will not be possible
as the typical rating is between 3 and 5 stars and the average will be calculated from substantially
more than ten ratings. Moreover, for this to be possible the user needs to watch in real time how
his average rating changes in order to associate the rating with a transaction through the classic
rating that is left at the same time and then shown explicitly in the feedback record.
20
For this we downloaded the listings of auctions that ended in the past. This can be done using
the ‘advanced search’ option on eBay. 59 users have listed an item in two categories and one user
in three categories. These multiple listings were treated as multiple observations in our data set.
21
In particular, we use the information on the number of ratings, by type, that were received within
the last month, respectively. They are added over the four waves. For the ratings that were
received in the last five months prior to the change we use the number of ratings received
in the “past six months” that was published on June 1 and subtract the number of ratings received
in the ‘last month’.
22
Table 4 shows percentiles of the distribution of feedback scores by availability of summary
statistics for detailed ratings. It reflects that these are only available if at least ten such ratings
have been left, i.e., for the more active sellers with higher feedback scores.
23
Figure 2 shows that the third quartile of the average rating per user is well below 5 stars.
24
Dellarocas and Wood (2008) investigate the consequences of non-random missing feedbacks
on feedback scores. They find that dissatisfied traders are more likely not to give feedback.
25
An additional advantage of these features is that separate reputation measures with respect to four
key criteria are calculated so that more detailed information is communicated to the community.
26
One might object that such a delay is costly because in the meantime an untrustworthy user may
deceive more eBay users. However, this would also happen if unsatisfied trading partners would
abstain from leaving a negative rating because they fear retaliation.
27
That is, both whether a feedback was left and the type of feedback should both be concealed to
opponents. This is somewhat different from the suggestion of concealing only the type of
feedback left which has been made in independent work by Reichling (2004).
28
This is also suitable for e-procurement platforms. See Dini and Spagnolo (2006) and
Dellarocas et al. (2006) for further details.